tqcq/llama.cpp
0
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2025-06-29 12:35:16 +00:00
Code Issues Packages Projects Releases Wiki Activity

vulkan: implement several ops relevant for ggml_opt (#11769)

Browse Source 
* vulkan: support memset_tensor

* vulkan: support GGML_OP_SUM

* vulkan: implement GGML_OP_ARGMAX

* vulkan: implement GGML_OP_SUB

* vulkan: implement GGML_OP_COUNT_EQUAL

* vulkan: implement GGML_OP_OPT_STEP_ADAMW

* vulkan: fix check_results RWKV_WKV6 crash and memory leaks

* vulkan: implement GGML_OP_REPEAT_BACK

* tests: remove invalid test-backend-ops REPEAT_BACK tests

* vulkan: fix COUNT_EQUAL memset using a fillBuffer command
This commit is contained in:
Rémy O
2025-02-17 07:55:57 +01:00
committed by GitHub
parent 0f2bbe6564
commit 2eea03d86a
8 changed files with 569 additions and 223 deletions
Download Patch File Download Diff File Expand all files Collapse all files

10
tests/test-backend-ops.cpp
View File

@ -1254,7 +1254,7 @@ struct test_count_equal : public test_case {
ggml_tensor * b = ggml_new_tensor(ctx, type, 4, ne.data());
ggml_set_name(b, "b");
ggml_tensor * b_argmax = ggml_argmax(ctx, a);
ggml_tensor * b_argmax = ggml_argmax(ctx, b);
ggml_set_name(b_argmax, "b_argmax");
ggml_tensor * out = ggml_count_equal(ctx, a_argmax, b_argmax);
@ -1511,6 +1511,7 @@ struct test_cont : public test_case {
};
// GGML_OP_ADD
// GGML_OP_SUB
// GGML_OP_MUL
// GGML_OP_DIV
struct test_bin_bcast : public test_case {
@ -3860,7 +3861,8 @@ static std::vector<std::unique_ptr<test_case>> make_test_cases_eval() {
test_cases.emplace_back(new test_conv_transpose_1d({3,2,1,1}, {3,1,2,1}, 1, 0, 1));
test_cases.emplace_back(new test_conv_transpose_1d({2,1,1,1}, {3,1,1,1}, 1, 0, 1));
test_cases.emplace_back(new test_count_equal());
test_cases.emplace_back(new test_count_equal(GGML_TYPE_F32, {4, 500, 1, 1}));
test_cases.emplace_back(new test_count_equal(GGML_TYPE_F32, {4, 5000, 1, 1}));
test_cases.emplace_back(new test_argmax(GGML_TYPE_F32, {32, 1, 1, 1}));
test_cases.emplace_back(new test_argmax(GGML_TYPE_F32, {100, 10, 1, 1}));
@ -3885,8 +3887,6 @@ static std::vector<std::unique_ptr<test_case>> make_test_cases_eval() {
test_cases.emplace_back(new test_repeat_back(GGML_TYPE_F32, {8, 6, 4, 2}, {1, 2, 1, 1}, view));
test_cases.emplace_back(new test_repeat_back(GGML_TYPE_F32, {8, 6, 4, 2}, {1, 1, 2, 1}, view));
test_cases.emplace_back(new test_repeat_back(GGML_TYPE_F32, {8, 6, 4, 2}, {1, 1, 1, 2}, view));
test_cases.emplace_back(new test_repeat_back(GGML_TYPE_I32, {8, 6, 4, 2}, {2, 1, 1, 1}, view));
test_cases.emplace_back(new test_repeat_back(GGML_TYPE_I16, {8, 6, 4, 2}, {1, 1, 1, 2}, view));
}
test_cases.emplace_back(new test_dup(GGML_TYPE_F32));
@ -3938,7 +3938,7 @@ static std::vector<std::unique_ptr<test_case>> make_test_cases_eval() {
test_cases.emplace_back(new test_cont(GGML_TYPE_BF16, {2, 3, 5 ,7}));
auto add_test_bin_bcast = [&](ggml_type type, std::array<int64_t, 4> ne, std::array<int, 4> nr) {
for (auto op : {ggml_add, ggml_mul, ggml_div}) {
for (auto op : {ggml_add, ggml_sub, ggml_mul, ggml_div}) {
test_cases.emplace_back(new test_bin_bcast(op, type, ne, nr));
}
};